LaszloKecskes
commented
3 months ago With the help of @gabriel-piles, we were unable to reproduce the error on staging. We have:
- initiated a fresh, full segmentation of a staging instance with a lot of files
- through the few errors we have identified the specific error type from @gabriel-piles' notes and every example was correct.
Locally, the error was caused by a faulty setup: accidentally, multiple tenants were pointing to the same database, and the process picked up the same new files twice for the two tenant entries.
Closing the issue without code changes, to be reopened if we see it again in staging.
The segmentation process that runs after uploading a pdf puts two segmentation entries in to the database. After it receives the results from the service, it updates only one of those entries, and one will have one correct entry and one stuck in 'processing' state. Under certain circumstances this makes the IX functionality unusable, as the system does not pick up the correct segmentation.
Manually deleting the faulty entries (leaving the one correct entry) is a work around. A migration is not necessary, as to 'processing' entries have a TTL of 24 hours.
This is put directly to sprint and worked on, since it blocks the priority IX development (relationships support), and is already present in staging.